# DECISIONS.md — Architecture Decision Log

Record any choice that deviates from or fills a gap in the docs. Newest at the bottom.

Format:
```
## D-NNN: Title
- Date:
- Context: (what problem / ambiguity)
- Decision: (what you chose)
- Alternatives considered:
- Consequences:
```

## D-001: SQLite + better-sqlite3 as the datastore
- Date: (project start)
- Context: Needs persistent accounts and game state with zero setup for a local build.
- Decision: SQLite file in `data/`, synchronous driver, WAL mode.
- Alternatives considered: Postgres (heavier setup), JSON files (no integrity/transactions).
- Consequences: Single-process deployment; synchronous transactions give simple double-spend protection. Migrate to Postgres only if scaling beyond one server.

## D-002: Lazy time evaluation instead of tick loops
- Date: (project start)
- Context: Game must progress while players are offline.
- Decision: Store timestamps; compute energy/timers on demand.
- Consequences: No background workers, trivially testable with injected clocks.

## D-003: Vanilla frontend, no build step
- Date: (project start)
- Context: Reduce tooling surface area for a local-model build.
- Decision: Static HTML/CSS/JS served by Express.
- Consequences: Slightly more DOM boilerplate; far fewer moving parts.

## D-004: Both combatants attack every round
- Date: 2026-10-20 (review)
- Context: Combat treated "higher speed acts first" as "only the faster side acts", so the slower side never attacked.
- Decision: Each round, both attack; faster first; hero wins speed ties; a kill ends the round.
- Consequences: Round log has up to 2 entries per round (max 40). Win rates need re-tuning (see STATUS issue 1).

## D-022: Material sell prices use item-level formula
- Context: BALANCE.md says materials sell at `common 2×level, uncommon 4×level, rare 8×level` — but "level" was ambiguous (hero level vs item level). The code used `defaultRate * heroLevel / itemMinLevel`, making high-level heroes overpay and low-level heroes underpay for high-level materials.
- Decision: Use `materialSellPrice(rarity, itemMinLevel)` from `src/game/rewards.js` — price depends on the ITEM's level, not the hero's. BALANCE.md table corrected to match the formula (e.g., Drake Scale: 90g → 112g).
- Consequences: Existing shop tests that asserted hero-level prices updated; material sell prices are now consistent regardless of hero level.

## D-005: Equipment stats are counted once in derived stats
- Context: maxHp added equipment VIT twice (once in totalVit, once as a separate term).
- Decision: `calcDerived` in `src/game/stats.js` is the ONLY derived-stat formula; combat and services call it.

## D-006: Real SQLite session store everywhere; unref its cleanup timer
- Context: better-sqlite3-session-store ignores `expired.clear:false` (`clear || true`), so its setInterval kept tests alive; the project had used a memory store in tests, so the real store was never exercised.
- Decision: Wrap store creation so its interval is unref'd. Tests now use the real store.

## D-007: `src/utils/time.js` is the only clock
- Decision: Services call `time.now()`; tests use `time.setNow()`. Game functions take `nowMs`/`rng` as parameters.

## D-023: Shared grantItem for inventory item granting
- Date: 2026-10-23
- Context: `huntService` and `shopService` both need to grant items to inventory. `huntService` had a local `grantItem` that correctly creates one row per unit for non-stackable items, but `shopService` duplicated the logic incorrectly (inserted full quantity for non-stackables).
- Decision: Extract `grantItem(s, characterId, item, quantity, now)` to `src/game/inventory.js`. Both services import it. The service's stmtCache must include `stackRow`, `addToStack`, `insertInv` with the exact same SQL used by huntService.
- Consequences: Single source of truth for item-granting logic. Any future service that grants items (expeditions, quests, etc.) can reuse it.

<!-- Add D-008 onward below -->